I'd like to contribute from the electronic system design perspective, since that's what I studied in University, back when the original IBM Pc was being intrduced and most of work in multiprocessing was done on the System 370 mainframe. The principles are still valid here, and don't forget that software is part of the system.
My first thought, it that with respect each of you probably has different systems, so the benchmarks are hazy. We large systems people are ALWAYS sceptical of benchmarks, because of limited relevance. I'd love to see someone with the discipline to test a range of configurations on the same machine (where possible). This could be a mac Pro with dual Xeons, max memory & disk in Raid 5 (I'll explain why shortly) and a Windows OS if chips & cores could be selectively disabled while memory, disk, are constant.
Second, any system configuration is bottleneck bound. I have done both business & engineering IT and found that engineers are generally choked by memory bandwidth, while business is choked by I/O bandwidth. I can often see which is causing me troblu on my PC without any performance monitoring tools, and I specified the Mac as I did to hopefully remove memory & I/O bottlenecks for truer CPU performance comparisons.
Now, my expectations are that multiple threaded, mutiple cored and multiple CPU system will show differing performance levels, and they may not be predictable. For example, if the above system were constrained to 1 CPU & 1 thread, a given application will have X performance. It might still only have X performance if the system were fully unconstrained because either the software cannot utilize the extra resources, or both configurations hit a memory or I/O bottleneck, it is ALWAYS application dependent. Another application COULD jump to say 32X performance in the switch 1 CPU, 1 Thread, to 2 CPU, 8 Core, 32 Thread (example, maybe not fact) because it doesn't overwhelm either memory or I/O, yet another run with different parameters could hit a chokepoint and come at much less than 32X.
This is all basic theory, which to same may be well known; but it also appears from your comments that is has been overlooked by many. My point is very simple, what you have been discussing has no absolute answers yet, because it all depends on so many variables.
Now, I found this post in doing research for a friend. It doesn't answer my question though: where can I find a current list of games and other software that can justify the purchase of a quad-core PC over a dual-core PC.